Medical record automatic abstract model generation method and device and computer equipment

Abstract

The invention relates to a medical record automatic abstract model generation method and device, computer equipment and a storage medium. The method comprises the following steps: performing test processing on each candidate student model according to each optimal abstract cue word and each optimal evaluation cue word to obtain a test result of the corresponding candidate student model, and determining an optimal student model according to each test result; the optimal student model is a candidate student model corresponding to the optimal test result; and obtaining a training medical record sample set, training the optimal student model according to the training medical record sample set, each optimal abstract cue word and each optimal evaluation cue word, and adjusting model parameters of the optimal student model based on an evaluation score index corresponding to the optimal evaluation cue word obtained by training and testing. And generating a trained medical record automatic abstract model until the evaluation score of each optimal evaluation prompt word meets a preset requirement, and outputting an optimal medical record abstract based on the trained medical record automatic abstract model. By adopting the method, the abstract efficiency and accuracy can be improved.

Description

Technical Field

[0001] This application relates to the technical field of natural language text processing, and particularly to a method, apparatus, computer device, and storage medium for generating a medical record automatic summary model. Background Art

[0002] A medical record summary is a concise and comprehensive summary of a patient's medical record information. With the development of medical technology and the increase in patient data volume, doctors and medical workers need to process more and more information. Traditional medical record automatic summary technologies generally fall into two categories. One is the extraction-based scheme, that is, extracting key fragments from the original text as the summary result; the other is the generation-based scheme, that is, generating a new summary of the original text content based on understanding the original text. Among them, the generation-based scheme is more reasonable intuitively, but in the early years, it was restricted by the model architecture and computing power, so it was on par with the extraction-based scheme. However, with the rapid improvement of large models led by GPT in recent years, the generation-based scheme has gradually become the mainstream.

[0003] However, traditional medical record automatic summary technologies have problems such as low efficiency or low accuracy. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, apparatus, computer device, and storage medium for generating a medical record automatic summary model that can improve the summary efficiency and accuracy.

[0005] In a first aspect, a method for generating a medical record automatic summary model is provided, and the method includes:

[0006] Obtain each medical record type, the corresponding candidate summary prompt word set for the medical record type, and the corresponding candidate scoring prompt word set for the medical record type; wherein, the medical record type includes outpatient and emergency medical records or inpatient medical records; the candidate summary prompt word set includes each candidate summary prompt word; the candidate scoring prompt word set includes each candidate scoring prompt word;

[0007] Perform summary effect evaluation processing on each candidate summary prompt word set and each candidate scoring prompt word set to obtain the optimal summary prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type;

[0008] Test each candidate student model according to each optimal summary prompt word and each optimal evaluation prompt word, and obtain the test results of the corresponding candidate student models. Determine the optimal student model according to each test result; the optimal student model is the candidate student model corresponding to the optimal test result;

[0009] Obtain a training medical record sample set, train an optimal student model based on the training medical record sample set, each optimal summary prompt word, and each optimal evaluation prompt word, and adjust the model parameters of the optimal student model based on the evaluation score indicators corresponding to the optimal evaluation prompt words obtained from training and testing until the evaluation scores of each optimal evaluation prompt word meet the preset requirements, generating a trained medical record automatic summary model to output an optimal medical record summary based on the trained medical record automatic summary model.

[0010] In one embodiment, obtaining each medical record type, the corresponding candidate summary prompt word set of the medical record type, and the corresponding candidate scoring prompt word set of the medical record type includes: obtaining the optimized artificial summary prompt words of each medical record type; in response to receiving the first natural language description information of each medical record type, inputting each optimized scoring prompt word and the first natural language description information of the corresponding medical record type into a teacher model for multiple rounds of summary prompt word rewriting processing to obtain the candidate summary prompt words of the corresponding medical record type; the first natural language description information is used to instruct the teacher model to perform summary prompt word rewriting processing; inputting the optimized artificial summary prompt words of each medical record type and the candidate summary prompt words of the corresponding medical record type into the empty set of candidate summary prompt words of the corresponding medical record type to obtain the candidate summary prompt word set of the corresponding medical record type.

[0011] In one embodiment, obtaining the optimized artificial summary prompt words of each medical record type includes: in response to receiving the initial artificial summary prompt words of each medical record type and the original medical records of the corresponding medical record types, inputting each initial artificial summary prompt word and the original medical records of the corresponding medical record types into a teacher model for medical record automatic summary processing to obtain the medical record summaries of the corresponding medical record types; in response to performing a first prompt word optimization operation on the initial artificial summary prompt words of each medical record type according to the medical record summaries of the corresponding medical record types, determining the optimized artificial summary prompt words of the corresponding medical record types according to the results of the first prompt word optimization operation.

[0012] In one embodiment, obtaining each medical record type, the corresponding candidate summary prompt word set of the medical record type, and the corresponding candidate scoring prompt word set of the medical record type further includes: obtaining the optimized artificial scoring prompt words of each medical record type; in response to receiving the second natural language description information of each medical record type, inputting each optimized artificial scoring prompt word and the second natural language description information of the corresponding medical record type into a teacher model for multiple rounds of scoring prompt word rewriting processing to obtain the candidate scoring prompt words of the corresponding medical record type; the second natural language description information is used to instruct the teacher model to perform scoring prompt word rewriting processing; inputting the optimized artificial scoring prompt words of each medical record type and the candidate scoring prompt words of the corresponding medical record type into the empty set of candidate scoring prompt words of the corresponding medical record type to obtain the candidate scoring prompt word set of the corresponding medical record type.

[0013] In one embodiment, obtaining the optimized manual scoring prompt words for each medical record type includes: in response to receiving the initial manual evaluation prompt words for each medical record type, the original medical records of the corresponding medical record types, and the medical record abstracts of the corresponding medical record types, inputting the initial manual evaluation prompt words, the original medical records of the corresponding medical record types, and the medical record abstracts of the corresponding medical record types into the teacher model for abstract scoring processing to obtain the first abstract scoring results for the corresponding medical record types; in response to performing a second prompt word optimization operation on the initial manual evaluation prompt words for the corresponding medical record types according to the first abstract scoring results for each medical record type, determining the optimized manual scoring prompt words for the corresponding medical record types according to the results of the second prompt word optimization operation.

[0014] In one embodiment, performing an abstract effect evaluation process on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type, including:

[0015] Sequentially performing a prompt word random extraction process on the candidate abstract prompt word sets for each medical record type to obtain the target candidate abstract prompt words for the corresponding medical record types;

[0016] Obtaining each target teacher model and the original medical records corresponding to each medical record type; the target teacher model is a teacher model configured and processed according to each preset model generation and each model parameter set;

[0017] Sequentially inputting the original medical records corresponding to each medical record type, the target candidate abstract prompt words for the corresponding medical record type, and the candidate scoring prompt words corresponding to the candidate scoring prompt word sets for the corresponding medical record type into each target teacher model to obtain a target teacher model abstract scoring result set of the candidate scoring prompt words corresponding to the candidate scoring prompt word sets for the corresponding medical record types; the target teacher model set includes the second abstract scoring results of the corresponding target teacher models;

[0018] Analyzing and comparing the target teacher model abstract scoring result set of the candidate scoring prompt words corresponding to the candidate scoring prompt word sets for each medical record type and a preset standard abstract scoring result set to obtain the optimal evaluation prompt word for the corresponding medical record type.

[0019] In one embodiment, performing an abstract effect evaluation process on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type, including:

[0020] Input the candidate abstract prompt words corresponding to each medical record type in the candidate abstract prompt word set corresponding to the corresponding medical record type, the optimal evaluation prompt words of the corresponding medical record type, and the original medical records corresponding to each medical record type into the teacher model in sequence to obtain the third abstract scoring result of the candidate abstract prompt words corresponding to the candidate abstract prompt word set corresponding to the corresponding medical record type;

[0021] Determine the optimal abstract prompt word of the corresponding medical record type according to the maximum value of the third abstract scoring result of the candidate abstract prompt words corresponding to the candidate abstract prompt word set corresponding to each medical record type.

[0022] In a second aspect, a device for generating an automatic medical record abstract model is provided. The device includes a prompt word set acquisition module, an abstract effect evaluation module, an optimal student model determination module, and a model training module.

[0023] Among them, the prompt word set acquisition module is used to acquire each medical record type, the candidate abstract prompt word set corresponding to the corresponding medical record type, and the candidate scoring prompt word set corresponding to the corresponding medical record type; among them, the medical record type includes emergency department medical records or inpatient medical records; the candidate abstract prompt word set includes each candidate abstract prompt word; the candidate scoring prompt word set includes each candidate scoring prompt word;

[0024] The abstract effect evaluation module is used to perform abstract effect evaluation processing on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word of the corresponding medical record type and the optimal evaluation prompt word of the corresponding medical record type; the optimal student model determination module is used to test each candidate student model according to each optimal abstract prompt word and each optimal evaluation prompt word to obtain the test result of the corresponding candidate student model, and determine the optimal student model according to each test result; the optimal student model is the candidate student model corresponding to the optimal test result; the model training module is used to acquire a training medical record sample set, train the optimal student model according to the training medical record sample set, each optimal abstract prompt word, and each optimal evaluation prompt word, and adjust the model parameters of the optimal student model based on the evaluation score index of the corresponding optimal evaluation prompt word obtained from training and testing until the evaluation scores of each optimal evaluation prompt word meet the preset requirements, and generate a trained automatic medical record abstract model to output an optimal medical record abstract based on the trained automatic medical record abstract model.

[0025] In a third aspect, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of any one of the methods in the above method embodiments are implemented.

[0026] In a fourth aspect, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of any one of the methods in the above method embodiments are implemented.

[0027] The above-mentioned method, device, computer equipment and storage medium for generating a medical record automatic summary model obtain various medical record types, a candidate summary prompt word set corresponding to the medical record type, and a candidate scoring prompt word set corresponding to the medical record type; wherein, the medical record types include outpatient and emergency medical records or inpatient medical records; the candidate summary prompt word set includes each candidate summary prompt word; the candidate scoring prompt word set includes each candidate scoring prompt word; then, perform a summary effect evaluation process on each candidate summary prompt word set and each candidate scoring prompt word set to obtain the optimal summary prompt word corresponding to the medical record type and the optimal evaluation prompt word corresponding to the medical record type; then, perform a test process on each candidate student model according to each optimal summary prompt word and each optimal evaluation prompt word to obtain the test result of the corresponding candidate student model, and determine the optimal student model according to each test result; the optimal student model is the candidate student model corresponding to the optimal test result; finally, obtain a training medical record sample set, train the optimal student model according to the training medical record sample set, each optimal summary prompt word and each optimal evaluation prompt word, and adjust the model parameters of the optimal student model based on the evaluation score index corresponding to the optimal evaluation prompt word obtained from training and testing until the evaluation scores of each optimal evaluation prompt word meet the preset requirements, and generate a trained medical record automatic summary model, so as to output the optimal medical record summary based on the trained medical record automatic summary model, which can improve the summary efficiency and accuracy, thereby facilitating the automatic summary of electronic medical records through the trained medical record automatic summary model, improving the summary efficiency and reducing the labor cost of writing and reading summaries. Summarizing through pathological type classification facilitates improving the accuracy of summary prompt words and evaluation prompt words, and further improves the comprehensiveness, objectivity and accuracy of the trained medical record automatic summary model. Description of the Drawings

[0028] Figure 1 It is an application environment diagram of the method for generating a medical record automatic summary model in an embodiment;

[0029] Figure 2 It is a schematic flowchart of the method for generating a medical record automatic summary model in an embodiment;

[0030] Figure 3 It is a first schematic flowchart of obtaining various medical record types, a candidate summary prompt word set corresponding to the medical record type, and a candidate scoring prompt word set corresponding to the medical record type in an embodiment;

[0031] Figure 4 It is a schematic flowchart of obtaining optimized manual summary prompt words for various medical record types in an embodiment;

[0032] Figure 5 It is a second schematic flowchart of obtaining various medical record types, a candidate summary prompt word set corresponding to the medical record type, and a candidate scoring prompt word set corresponding to the medical record type in an embodiment;

[0033] Figure 6 It is a schematic flow chart of obtaining optimized manual scoring prompt words for each medical record type in an embodiment;

[0034] Figure 7 It is a schematic flow chart of performing abstract effect evaluation processing on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type in an embodiment;

[0035] Figure 8 It is a structural block diagram of a medical record automatic abstract model generation device in an embodiment;

[0036] Figure 9 It is an internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0037] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0038] To facilitate the understanding of the present application, the present application will be described more comprehensively below with reference to the relevant accompanying drawings. Embodiments of the present application are shown in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used in the specification of the present application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application.

[0040] It can be understood that the terms "first", "second", etc. used in the present application can be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the present application, the first resistor can be called the second resistor, and similarly, the second resistor can be called the first resistor. Both the first resistor and the second resistor are resistors, but they are not the same resistor.

[0041] It can be understood that "connection" in the following embodiments should be understood as "electrical connection", "communication connection", etc. if there is an electrical signal or data transmission between the connected circuits, modules, units, etc.

[0042] As used herein, the singular forms "a", "an" and "the" may also include the plural forms unless the context clearly dictates otherwise. It should also be understood that the terms "comprises / comprising" or "has / have" etc. specify the presence of the stated features, integers, steps, operations, components, parts, or combinations thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof.

[0043] The method for generating an automatic medical record abstract model provided by this application can be applied to, for example, Figure 1 the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. Among them, the terminal 102 can be but is not limited to various personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices, and the server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0044] In one embodiment, as Figure 2 shown, a method for generating an automatic medical record abstract model is provided. Taking the server 104 in Figure 1 as an example for illustration, it includes the following steps 201 to step 204.

[0045] Step 201, obtain each medical record type, the corresponding candidate abstract prompt word set for the medical record type, and the corresponding candidate scoring prompt word set for the medical record type.

[0046] Among them, the medical record types include emergency department medical records or inpatient medical records; the candidate abstract prompt word set includes each candidate abstract prompt word; the candidate scoring prompt word set includes each candidate scoring prompt word; specifically, the server 104 obtains each medical record type, the corresponding candidate abstract prompt word set for the medical record type, and the corresponding candidate scoring prompt word set for the medical record type.

[0047] In one of the embodiments, as Figure 3 shown, obtaining each medical record type, the corresponding candidate abstract prompt word set for the medical record type, and the corresponding candidate scoring prompt word set for the medical record type includes steps 301 to step 303.

[0048] Step 301, obtain the optimized artificial abstract prompt words for each medical record type.

[0049] Specifically, the server 104 obtains the optimized artificial abstract prompt words for each medical record type.

[0050] In one of the embodiments, as Figure 4 shown, obtaining the optimized artificial abstract prompt words for each medical record type includes steps 401 to step 402.

[0051] Step 401: In response to receiving the initial manual abstract prompts for each medical record type and the original medical records of the corresponding medical record types, input the initial manual abstract prompts and the original medical records of the corresponding medical record types into the teacher model for automatic medical record abstract processing to obtain the medical record abstracts of the corresponding medical record types.

[0052] Step 402: In response to performing a first prompt optimization operation on the initial manual abstract prompts of the corresponding medical record types based on the medical record abstracts of the corresponding medical record types, determine the optimized manual abstract prompts of the corresponding medical record types according to the results of the first prompt optimization operation.

[0053] Among them, the initial manual abstract prompts are manually written and selected by doctors according to the corresponding medical record types. The teacher model is any large language model (LLM). In a specific example, the teacher model can be, but is not limited to, the GPT model. The above is only a specific example, and in actual applications, it can be flexibly set according to user needs and will not be restricted here.

[0054] In a specific example, the teacher model plays a crucial role in the model distillation technology. Model distillation is a knowledge transfer technology aimed at transferring the knowledge learned by a complex and usually high-performance teacher model to a relatively simple and smaller-scale student model. The teacher model is usually a powerful and complex model that can provide high-quality prediction results, but its running efficiency and resource consumption are relatively high. The student model is the target model in the knowledge distillation process, and its role is to improve the model compression and performance optimization by learning the behavior and output of the teacher model. The student model usually has a smaller number of parameters and computational complexity, but its performance needs to be improved by learning the knowledge of the teacher model. The above is only a specific example, and in actual applications, it can be flexibly set according to user needs and will not be restricted here.

[0055] Specifically, the server 104, in response to receiving the initial manual abstract prompts for each medical record type and the original medical records of the corresponding medical record types, inputs the initial manual abstract prompts and the original medical records of the corresponding medical record types into the teacher model for automatic medical record abstract processing to obtain the medical record abstracts of the corresponding medical record types; then, in response to performing a first prompt optimization operation on the initial manual abstract prompts of the corresponding medical record types based on the medical record abstracts of the corresponding medical record types, determines the optimized manual abstract prompts of the corresponding medical record types according to the results of the first prompt optimization operation, improving the accuracy and efficiency of the automatic medical record abstract model.

[0056] In this embodiment, in response to receiving the initial manual abstract prompt words for each medical record type and the original medical records of the corresponding medical record types, the initial manual abstract prompt words and the original medical records of the corresponding medical record types are input into the teacher model for automatic medical record abstract processing to obtain the medical record abstracts of the corresponding medical record types; then, in response to performing a first prompt word optimization operation on the initial manual abstract prompt words of the corresponding medical record types according to the medical record abstracts of each medical record type, the optimized manual abstract prompt words of the corresponding medical record types are determined according to the results of the first prompt word optimization operation, improving the accuracy and efficiency of the medical record automatic abstract model.

[0057] Step 302, in response to receiving the first natural language description information of each medical record type, the optimized scoring prompt words and the first natural language description information of the corresponding medical record types are input into the teacher model for multi-round abstract prompt word rewriting processing to obtain the candidate abstract prompt words of the corresponding medical record types.

[0058] Among them, the first natural language description information is used to instruct the teacher model to perform abstract prompt word rewriting processing. Specifically, the server 104, in response to receiving the first natural language description information of each medical record type, inputs the optimized scoring prompt words and the first natural language description information of the corresponding medical record types into the teacher model for multi-round abstract prompt word rewriting processing to obtain the candidate abstract prompt words of the corresponding medical record types.

[0059] Step 303, the optimized manual abstract prompt words of each medical record type and the candidate abstract prompt words of the corresponding medical record type are input into the empty set of candidate abstract prompt words of the corresponding medical record type to obtain the candidate abstract prompt word set of the corresponding medical record type.

[0060] Specifically, the server 104 inputs the optimized manual abstract prompt words of each medical record type and the candidate abstract prompt words of the corresponding medical record type into the empty set of candidate abstract prompt words of the corresponding medical record type to obtain the candidate abstract prompt word set of the corresponding medical record type, thereby improving the generation efficiency and convenience of the candidate abstract prompt word set, and also ensuring that the prompt words can fully stimulate the model's capabilities, so as to obtain an ideal medical record automatic abstract model.

[0061] In this embodiment, optimized manual abstract prompting words for each medical record type are obtained; then, in response to receiving the first natural language description information of each medical record type, the optimized scoring prompting words and the first natural language description information of the corresponding medical record type are input into the teacher model for multi-round abstract prompting word rewriting processing to obtain candidate abstract prompting words for the corresponding medical record type; then, the optimized manual abstract prompting words of each medical record type and the candidate abstract prompting words of the corresponding medical record type are input into the empty set of candidate abstract prompting words of the corresponding medical record type to obtain a set of candidate abstract prompting words for the corresponding medical record type, thereby improving the generation efficiency and convenience of the set of candidate abstract prompting words, and also ensuring that the prompting words can fully stimulate the model's ability, so as to obtain an ideal automatic medical record abstract model.

[0062] In one embodiment, as Figure 5 shown, obtaining each medical record type, the corresponding set of candidate abstract prompting words for the medical record type, and the corresponding set of candidate scoring prompting words for the medical record type further includes steps 501 to 502.

[0063] Step 501, obtaining optimized manual scoring prompting words for each medical record type.

[0064] In one embodiment, as Figure 6 shown, obtaining optimized manual scoring prompting words for each medical record type includes steps 601 and 602.

[0065] Step 601, in response to receiving the initial manual evaluation prompting words of each medical record type, the original medical record of the corresponding medical record type, and the medical record abstract of the corresponding medical record type, inputting the initial manual evaluation prompting words, the original medical record of the corresponding medical record type, and the medical record abstract of the corresponding medical record type into the teacher model for abstract scoring processing to obtain the first abstract scoring result of the corresponding medical record type;

[0066] Step 602, in response to performing a second prompting word optimization operation on the initial manual evaluation prompting words of the corresponding medical record type according to the first abstract scoring result of each medical record type, determining the optimized manual scoring prompting words of the corresponding medical record type according to the result of the second prompting word optimization operation.

[0067] Specifically, in response to receiving the initial manual evaluation prompt words for each medical record type, the original medical records of the corresponding medical record type, and the medical record abstracts of the corresponding medical record type, the server 104 inputs the initial manual evaluation prompt words, the original medical records of the corresponding medical record type, and the medical record abstracts of the corresponding medical record type into the teacher model for abstract scoring processing to obtain the first abstract scoring results for the corresponding medical record types; then, in response to performing a second prompt word optimization operation on the initial manual evaluation prompt words of the corresponding medical record types according to the first abstract scoring results of the respective medical record types, the optimized manual scoring prompt words for the corresponding medical record types are determined based on the results of the second prompt word optimization operation, improving the accuracy of the optimized manual scoring prompt words, and thus improving the accuracy and efficiency of the medical record automatic abstract model.

[0068] In this embodiment, in response to receiving the initial manual evaluation prompt words for each medical record type, the original medical records of the corresponding medical record type, and the medical record abstracts of the corresponding medical record type, the server 104 inputs the initial manual evaluation prompt words, the original medical records of the corresponding medical record type, and the medical record abstracts of the corresponding medical record type into the teacher model for abstract scoring processing to obtain the first abstract scoring results for the corresponding medical record types; then, in response to performing a second prompt word optimization operation on the initial manual evaluation prompt words of the corresponding medical record types according to the first abstract scoring results of the respective medical record types, the optimized manual scoring prompt words for the corresponding medical record types are determined based on the results of the second prompt word optimization operation, improving the accuracy of the optimized manual scoring prompt words, and thus improving the accuracy and efficiency of the medical record automatic abstract model.

[0069] Step 502, in response to receiving the second natural language description information for each medical record type, the optimized manual scoring prompt words and the second natural language description information of the corresponding medical record type are input into the teacher model for multi-round scoring prompt word rewriting processing to obtain the candidate scoring prompt words for the corresponding medical record types.

[0070] Step 503, the optimized manual scoring prompt words of each medical record type and the candidate scoring prompt words of the corresponding medical record type are input into the empty set of candidate scoring prompt words of the corresponding medical record type to obtain the candidate scoring prompt word set of the corresponding medical record type.

[0071] Among them, the second natural language description information is used to instruct the teacher model to perform rewrite processing on the scoring prompt words. Specifically, the server 104 obtains the optimized manual scoring prompt words for each medical record type. Then, in response to receiving the second natural language description information of each medical record type, the optimized manual scoring prompt words and the second natural language description information of the corresponding medical record type are input into the teacher model for multiple rounds of rewrite processing of the scoring prompt words, obtaining the candidate scoring prompt words for the corresponding medical record type; then, the optimized manual scoring prompt words of each medical record type and the candidate scoring prompt words of the corresponding medical record type are input into the empty set of candidate scoring prompt words of the corresponding medical record type, obtaining the candidate scoring prompt word set for the corresponding medical record type, thereby improving the generation efficiency and convenience of the candidate scoring prompt word set, and also ensuring that the prompt words can fully stimulate the model's ability, so as to obtain an ideal medical record automatic abstract model.

[0072] In this embodiment, the optimized manual scoring prompt words for each medical record type are obtained. Then, in response to receiving the second natural language description information of each medical record type, the optimized manual scoring prompt words and the second natural language description information of the corresponding medical record type are input into the teacher model for multiple rounds of rewrite processing of the scoring prompt words, obtaining the candidate scoring prompt words for the corresponding medical record type; then, the optimized manual scoring prompt words of each medical record type and the candidate scoring prompt words of the corresponding medical record type are input into the empty set of candidate scoring prompt words of the corresponding medical record type, obtaining the candidate scoring prompt word set for the corresponding medical record type, thereby improving the generation efficiency and convenience of the candidate scoring prompt word set, and also ensuring that the prompt words can fully stimulate the model's ability, so as to obtain an ideal medical record automatic abstract model.

[0073] Step 202, perform abstract effect evaluation processing on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type.

[0074] Specifically, the server 104 performs abstract effect evaluation processing on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type.

[0075] In one of the embodiments, as Figure 7 shown, performing abstract effect evaluation processing on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type includes steps 701 to 704.

[0076] Step 701, sequentially perform random extraction processing on the candidate abstract prompt word sets of each medical record type to obtain the target candidate abstract prompt words for the corresponding medical record type.

[0077] Step 702: Obtain the original medical records corresponding to each target teacher model and each medical record type.

[0078] Step 703: Sequentially input the original medical records corresponding to each medical record type, the target candidate abstract prompt words corresponding to the medical record type, and the candidate scoring prompt words in the corresponding candidate scoring prompt word set of the medical record type into each target teacher model respectively, to obtain the target teacher model abstract scoring result sets of the candidate scoring prompt words in the corresponding candidate scoring prompt word set of the medical record type.

[0079] Step 704: Analyze and compare according to the target teacher model abstract scoring result sets of the candidate scoring prompt words in the corresponding candidate scoring prompt word sets of each medical record type and the preset standard abstract scoring result sets, to obtain the optimal evaluation prompt words corresponding to the medical record type.

[0080] Among them, the target teacher model is the teacher model processed according to each preset model algebra and each model parameter set. The target teacher model set includes the second abstract scoring results of the corresponding target teacher models. The target teacher model set includes the second abstract scoring results of the corresponding target teacher models. Specifically, the server 104 sequentially performs prompt word random extraction processing on the candidate abstract prompt word sets of each medical record type to obtain the target candidate abstract prompt words corresponding to the medical record type; then, obtains the original medical records corresponding to each target teacher model and each medical record type; the target teacher model is the teacher model processed according to each preset model algebra and each model parameter set; then, sequentially input the original medical records corresponding to each medical record type, the target candidate abstract prompt words corresponding to the medical record type, and the candidate scoring prompt words in the corresponding candidate scoring prompt word set of the medical record type into each target teacher model respectively, to obtain the target teacher model abstract scoring result sets of the candidate scoring prompt words in the corresponding candidate scoring prompt word set of the medical record type; finally, analyze and compare according to the target teacher model abstract scoring result sets of the candidate scoring prompt words in the corresponding candidate scoring prompt word sets of each medical record type and the preset standard abstract scoring result sets, to obtain the optimal evaluation prompt words corresponding to the medical record type, which improves the efficiency and accuracy of determining the optimal evaluation prompt words and is convenient for evaluating the abstract ability of the model.

[0081] In a specific example, the selected teacher model is Model A, and the preset model generations include the Nth generation of Model A and the (N + 1)th generation of Model A. The model parameter sets include 7b parameters, 14b parameters, and 32b parameters. Therefore, the target teacher models include the first target teacher model, the second target teacher model, the third target teacher model, the fourth target teacher model, the fifth target teacher model, and the sixth target teacher model. The first target teacher model is the Nth generation of Model A configured with 7b parameters. The second target teacher model is the Nth generation of Model A configured with 14b parameters. The third target teacher model is the Nth generation of Model A configured with 32b parameters. The fourth target teacher model is the (N + 1)th generation of Model A configured with 7b parameters. The fifth target teacher model is the (N + 1)th generation of Model A configured with 14b parameters. The sixth target teacher model is the (N + 1)th generation of Model A configured with 32b parameters. Each candidate scoring prompt word and the corresponding original medical record of the corresponding medical record type, and the target candidate summary prompt word of the corresponding medical record type are sequentially selected from the candidate scoring prompt word set of the corresponding medical record type and input into each target teacher model to obtain the target teacher model summary scoring result set of the candidate scoring prompt word in the candidate scoring prompt word set of the corresponding medical record type. It can be understood that the target teacher model summary scoring result set includes the second summary scoring results of the first target teacher model, the second target teacher model, the third target teacher model, the fourth target teacher model, the fifth target teacher model, and the sixth target teacher model corresponding to the candidate scoring prompt word. The target teacher model summary scoring result set of the candidate scoring prompt word in the candidate scoring prompt word set of each medical record type is analyzed and compared with the preset standard summary scoring result set, and then it can be known whether the target teacher model summary scoring result set of the candidate scoring prompt word in the candidate scoring prompt word set of each medical record type has the largest score difference and conforms to the score ranking expectation of the standard summary scoring result set, and the candidate scoring prompt word corresponding to the second summary scoring result with the largest score difference and conforming to the score ranking of the standard summary scoring result set is determined as the optimal evaluation prompt word for the corresponding medical record type. The above is only a specific example, and it is flexibly set according to user needs in actual applications and is not limited here.

[0082] In this embodiment, random extraction processing is performed on the candidate abstract prompt word sets of each medical record type to obtain the target candidate abstract prompt words of the corresponding medical record type; then, the original medical records corresponding to each target teacher model and each medical record type are obtained; the target teacher model is the teacher model configured and processed according to each preset model algebra and each model parameter set; then, the original medical records corresponding to each medical record type, the target candidate abstract prompt words of the corresponding medical record type, and the candidate scoring prompt words corresponding to the candidate scoring prompt word sets of the corresponding medical record type are sequentially input into each target teacher model to obtain the target teacher model abstract scoring result sets of the candidate scoring prompt words corresponding to the candidate scoring prompt word sets of the corresponding medical record type; the target teacher model set includes the second abstract scoring results of the corresponding target teacher models; finally, analysis and comparison are performed according to the target teacher model abstract scoring result sets of the candidate scoring prompt words corresponding to the candidate scoring prompt word sets of each medical record type and the preset standard abstract scoring result sets to obtain the optimal evaluation prompt words of the corresponding medical record type, improving the efficiency and accuracy of determining the optimal evaluation prompt words and facilitating the evaluation of the abstract ability of the model.

[0083] In one embodiment, as Figure 7 shown, abstract effect evaluation processing is performed on each candidate abstract prompt word set and each candidate scoring prompt word set to obtain the optimal abstract prompt words of the corresponding medical record type and the optimal evaluation prompt words of the corresponding medical record type, including steps 705 to 706.

[0084] Step 705, sequentially input the original medical records corresponding to each medical record type, the optimal evaluation prompt words of the corresponding medical record type, and the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of the corresponding medical record type into the teacher model to obtain the third abstract scoring results of the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of the corresponding medical record type.

[0085] Step 706, determine the optimal abstract prompt words of the corresponding medical record type according to the maximum value of the third abstract scoring results of the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of each medical record type.

[0086] Specifically, the server 104 sequentially inputs the original medical records corresponding to each medical record type, the optimal evaluation prompt words of the corresponding medical record type, and the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of the corresponding medical record type into the teacher model to obtain the third abstract scoring results of the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of the corresponding medical record type; then, the optimal abstract prompt words of the corresponding medical record type are determined according to the maximum value of the third abstract scoring results of the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of each medical record type, improving the efficiency and accuracy of determining the optimal abstract prompt words and facilitating the improvement of the abstract ability of the model.

[0087] In this embodiment, the original medical records corresponding to each medical record type, the optimal evaluation prompt words of the corresponding medical record type, and the candidate abstract prompt words in the candidate abstract prompt word set corresponding to the corresponding medical record type are sequentially input into the teacher model to obtain the third abstract scoring results of the candidate abstract prompt words in the candidate abstract prompt word set corresponding to the corresponding medical record type; then, the optimal abstract prompt word of the corresponding medical record type is determined according to the maximum value of the third abstract scoring results of the candidate abstract prompt words in the candidate abstract prompt word set of each medical record type, which improves the efficiency and accuracy of determining the optimal abstract prompt word and is convenient for improving the abstract ability of the model.

[0088] Step 203: After testing each candidate student model according to each optimal abstract prompt word and each optimal evaluation prompt word, obtain the test results of the corresponding candidate student models, and determine the optimal student model according to each test result.

[0089] Among them, the optimal student model is the candidate student model corresponding to the optimal test result. The server 104 tests each candidate student model according to the optimal abstract prompt words of each medical record type and the optimal evaluation prompt words of the corresponding medical record type to obtain the test results of the corresponding candidate student models, and determines the optimal student model according to each test result.

[0090] Step 204: Obtain a training medical record sample set, train the optimal student model according to the training medical record sample set, each optimal abstract prompt word, and each optimal evaluation prompt word, and adjust the model parameters of the optimal student model based on the evaluation score index corresponding to the optimal evaluation prompt word obtained from training and testing until the evaluation scores of each optimal evaluation prompt word meet the preset requirements, and generate a trained medical record automatic abstract model to output the optimal medical record abstract based on the trained medical record automatic abstract model.

[0091] Specifically, the server 104 obtains a training medical record sample set, trains the optimal student model according to the training medical record sample set, each optimal abstract prompt word, and each optimal evaluation prompt word, and adjusts the model parameters of the optimal student model based on the evaluation score index corresponding to the optimal evaluation prompt word obtained from training and testing until the evaluation scores of each optimal evaluation prompt word meet the preset requirements, and generates a trained medical record automatic abstract model to output the optimal medical record abstract based on the trained medical record automatic abstract model, so as to facilitate the automatic abstract of electronic medical records through the trained medical record automatic abstract model, improve the abstract efficiency and reduce the labor cost of writing and reading abstracts. By classifying according to the pathological type for abstracting, it is convenient to improve the accuracy of the abstract prompt words and evaluation prompt words, and further improve the comprehensiveness, objectivity, and accuracy of the trained medical record automatic abstract model.

[0092] In the above method for generating a medical record automatic summary model, candidate summary prompt word sets corresponding to each medical record type, and candidate scoring prompt word sets corresponding to the medical record types are obtained; wherein, the medical record types include outpatient and emergency medical records or inpatient medical records; the candidate summary prompt word sets include each candidate summary prompt word; the candidate scoring prompt word sets include each candidate scoring prompt word; then, summary effect evaluation processing is performed on each candidate summary prompt word set and each candidate scoring prompt word set to obtain the optimal summary prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type; next, each candidate student model is tested according to each optimal summary prompt word and each optimal evaluation prompt word, and the test results of the corresponding candidate student models are obtained, and the optimal student model is determined according to each test result; the optimal student model is the candidate student model corresponding to the optimal test result; finally, a training medical record sample set is obtained, and the optimal student model is trained according to the training medical record sample set, each optimal summary prompt word, and each optimal evaluation prompt word, and the model parameters of the optimal student model are adjusted based on the evaluation score index corresponding to the optimal evaluation prompt word obtained from training and testing until the evaluation scores of each optimal evaluation prompt word meet the preset requirements, and a trained medical record automatic summary model is generated to output an optimal medical record summary based on the trained medical record automatic summary model, which can improve the summary efficiency and accuracy, thereby facilitating the automatic summary of electronic medical records through the trained medical record automatic summary model, improving the summary efficiency and reducing the labor cost of writing and reading summaries, and performing summaries through pathological type classification, which is convenient for improving the accuracy of summary prompt words and evaluation prompt words, and further improving the comprehensiveness, objectivity, and accuracy of the trained medical record automatic summary model.

[0093] In a specific example, obtaining a training medical record sample set includes:

[0094] Obtaining a preset number of original medical records for each medical record type;

[0095] Inputting the preset number of original medical records for each medical record type, the optimal summary prompt word corresponding to the medical record type, and the optimal evaluation prompt word corresponding to the medical record type into a teacher model respectively to obtain a medical record summary sample corresponding to the medical record type;

[0096] Generating a training medical record sample set according to the preset number of original medical records for each medical record type, the optimal summary prompt word corresponding to the medical record type, the optimal evaluation prompt word corresponding to the medical record type, and the medical record summary sample corresponding to the medical record type. The above is only a specific example, and in actual applications, it can be flexibly set according to user needs and will not be limited here.

[0097] In a specific example, the method further includes:

[0098] Responding to an input operation on an electronic medical record to determine the electronic medical record;

[0099] Determine the corresponding medical record type according to the electronic medical record;

[0100] Input the electronic medical record and the corresponding medical record type into the trained medical record automatic summarization model to output the optimal medical record summary. The above is only a specific example, and in actual applications, it can be flexibly set according to user needs and is not restricted here.

[0101] It should be understood that although Figure 2-7 each step in the flowchart of Figure 2-7 is shown in sequence according to the indication of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless there is a clear description in this article, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover,

[0102] In a second aspect, as Figure 8 shown, a device for generating a medical record automatic summarization model is provided. The device includes a prompt word set acquisition module 810, a summary effect evaluation module 820, an optimal student model determination module 830, and a model training module 840.

[0103] Among them, the prompt set acquisition module 810 is used to acquire candidate abstract prompt sets and candidate scoring prompt sets corresponding to each medical record type; among them, the medical record types include emergency department medical records or inpatient medical records; the candidate abstract prompt sets include each candidate abstract prompt; the candidate scoring prompt sets include each candidate scoring prompt; the abstract effect evaluation module 820 is used to perform abstract effect evaluation processing on each candidate abstract prompt set and each candidate scoring prompt set to obtain the optimal abstract prompt and the optimal evaluation prompt corresponding to the corresponding medical record type; the optimal student model determination module 830 is used to test each candidate student model according to each optimal abstract prompt and each optimal evaluation prompt to obtain the test results of the corresponding candidate student model, and determine the optimal student model according to each test result; the optimal student model is the candidate student model corresponding to the optimal test result; the model training module 840 is used to acquire a training medical record sample set, train the optimal student model according to the training medical record sample set, each optimal abstract prompt and each optimal evaluation prompt, and adjust the model parameters of the optimal student model based on the evaluation score index corresponding to the trained and tested optimal evaluation prompt until the evaluation scores of each optimal evaluation prompt meet the preset requirements, and generate a trained medical record automatic abstract model to output the optimal medical record abstract based on the trained medical record automatic abstract model.

[0104] In one embodiment, the prompt set acquisition module 810 includes an abstract prompt set acquisition unit.

[0105] Among them, the abstract prompt set acquisition unit is used to acquire the optimized artificial abstract prompts for each medical record type; the abstract prompt set acquisition unit is used to, in response to receiving the first natural language description information of each medical record type, input each optimized scoring prompt and the first natural language description information of the corresponding medical record type into the teacher model for multiple rounds of abstract prompt rewriting processing to obtain the candidate abstract prompts corresponding to the corresponding medical record type; the first natural language description information is used to instruct the teacher model to perform abstract prompt rewriting processing; the abstract prompt set acquisition unit is used to input the optimized artificial abstract prompts of each medical record type and the candidate abstract prompts corresponding to the corresponding medical record type into the empty set of candidate abstract prompts of the corresponding medical record type to obtain the candidate abstract prompt set corresponding to the corresponding medical record type.

[0106] In one embodiment, the abstract prompt set acquisition unit further includes an optimized artificial abstract prompt acquirer.

[0107] Among them, the optimized artificial abstract prompt word acquirer is used to, in response to receiving the initial artificial abstract prompt words of each medical record type and the original medical records of the corresponding medical record types, input the initial artificial abstract prompt words and the original medical records of the corresponding medical record types into the teacher model for automatic medical record abstract processing to obtain the medical record abstracts of the corresponding medical record types; the optimized artificial abstract prompt word acquirer is used to, in response to performing a first prompt word optimization operation on the initial artificial abstract prompt words of the corresponding medical record types according to the medical record abstracts of the corresponding medical record types, determine the optimized artificial abstract prompt words of the corresponding medical record types according to the results of the first prompt word optimization operation.

[0108] In one embodiment, the prompt word set acquisition module 810 further includes a scoring prompt word set acquisition unit.

[0109] Among them, the scoring prompt word set acquisition unit is used to acquire the optimized artificial scoring prompt words of each medical record type; the scoring prompt word set acquisition unit is used to, in response to receiving the second natural language description information of each medical record type, input the optimized artificial scoring prompt words and the second natural language description information of the corresponding medical record types into the teacher model for multi-round scoring prompt word rewriting processing to obtain the candidate scoring prompt words of the corresponding medical record types; the second natural language description information is used to instruct the teacher model to perform scoring prompt word rewriting processing; the scoring prompt word set acquisition unit is used to input the optimized artificial scoring prompt words and the candidate scoring prompt words of the corresponding medical record types into the empty set of candidate scoring prompt words of the corresponding medical record types to obtain the candidate scoring prompt word sets of the corresponding medical record types.

[0110] In one embodiment, the scoring prompt word set acquisition unit includes an optimized artificial scoring prompt word acquirer.

[0111] Among them, the optimized artificial scoring prompt word acquirer is used to, in response to receiving the initial artificial evaluation prompt words of each medical record type, the original medical records of the corresponding medical record types, and the medical record abstracts of the corresponding medical record types, input the initial artificial evaluation prompt words, the original medical records of the corresponding medical record types, and the medical record abstracts of the corresponding medical record types into the teacher model for abstract scoring processing to obtain the first abstract scoring results of the corresponding medical record types; the optimized artificial scoring prompt word acquirer is used to, in response to performing a second prompt word optimization operation on the initial artificial evaluation prompt words of the corresponding medical record types according to the first abstract scoring results of the corresponding medical record types, determine the optimized artificial scoring prompt words of the corresponding medical record types according to the results of the second prompt word optimization operation.

[0112] In one embodiment, the abstract effect evaluation module 820 includes an abstract effect evaluation unit.

[0113] Among them, the abstract effect evaluation unit is used to perform random extraction processing on the candidate abstract prompt word sets of each medical record type in sequence to obtain the target candidate abstract prompt words of the corresponding medical record type;

[0114] Obtain the original medical records corresponding to each target teacher model and each medical record type; the target teacher model is the teacher model configured and processed according to each preset model generation and each model parameter set; the abstract effect evaluation unit is used to sequentially input the original medical records corresponding to each medical record type, the target candidate abstract prompt words of the corresponding medical record type, and the candidate scoring prompt words corresponding to the candidate scoring prompt word sets of the corresponding medical record type into each target teacher model to obtain the target teacher model abstract scoring result sets of the candidate scoring prompt words corresponding to the candidate scoring prompt word sets of the corresponding medical record type; the target teacher model set includes the second abstract scoring results of the corresponding target teacher models; the abstract effect evaluation unit is used to perform analysis and comparison based on the target teacher model abstract scoring result sets of the candidate scoring prompt words corresponding to the candidate scoring prompt word sets of each medical record type and the preset standard abstract scoring result sets to obtain the optimal evaluation prompt words of the corresponding medical record type.

[0115] In one embodiment, the abstract effect evaluation unit is used to sequentially input the original medical records corresponding to each medical record type, the optimal evaluation prompt words of the corresponding medical record type, and the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of the corresponding medical record type into the teacher model to obtain the third abstract scoring results of the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of the corresponding medical record type; the abstract effect evaluation unit is used to determine the optimal abstract prompt words of the corresponding medical record type according to the maximum value of the third abstract scoring results of the candidate abstract prompt words corresponding to the candidate abstract prompt word sets of each medical record type.

[0116] For the specific limitations of the medical record automatic abstract model generation device, reference can be made to the limitations of the medical record automatic abstract model generation method in the above text, which will not be elaborated here. Each module in the above medical record automatic abstract model generation device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor in the computer device in hardware form or independent of it, or stored in the memory in the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0117] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 9As shown in the figure. The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store various medical record types, the candidate summary prompt word sets corresponding to the medical record types, and the candidate scoring prompt word sets corresponding to the medical record types. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a method for generating an automatic medical record summary model.

[0118] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0119] In a third aspect, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the steps of any one of the methods in the above method embodiments.

[0120] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. When the computer program is executed by the processor, it implements the steps of any one of the methods in the above method embodiments.

[0121] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0122] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0123] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A method for generating a medical record automatic abstract model, the method comprising: Obtaining each medical record type, the corresponding candidate abstract prompt word set for the medical record type, and the corresponding candidate scoring prompt word set for the medical record type; wherein, the medical record type includes outpatient and emergency medical records or inpatient medical records; the candidate abstract prompt word set includes each candidate abstract prompt word; the candidate scoring prompt word set includes each candidate scoring prompt word; Performing an abstract effect evaluation process on each of the candidate abstract prompt word sets and each of the candidate scoring prompt word sets to obtain the optimal abstract prompt word for the corresponding medical record type and the optimal evaluation prompt word for the corresponding medical record type; Testing each candidate student model according to each of the optimal abstract prompt words and each of the optimal evaluation prompt words to obtain the test results of the corresponding candidate student models, and determining the optimal student model according to each of the test results; the optimal student model is the candidate student model corresponding to the optimal test result; Obtaining a training medical record sample set, training the optimal student model according to the training medical record sample set, each of the optimal abstract prompt words, and each of the optimal evaluation prompt words, and adjusting the model parameters of the optimal student model based on the evaluation score index corresponding to the optimal evaluation prompt word obtained from training and testing until the evaluation scores of each of the optimal evaluation prompt words meet the preset requirements, generating the trained medical record automatic abstract model, so as to output the optimal medical record abstract based on the trained medical record automatic abstract model.

2. The method according to claim 1, wherein The obtaining each medical record type, the corresponding candidate abstract prompt word set for the medical record type, and the corresponding candidate scoring prompt word set for the medical record type includes: Obtaining the optimized artificial abstract prompt words for each of the medical record types; In response to receiving the first natural language description information for each of the medical record types, inputting each of the optimized scoring prompt words and the first natural language description information for the corresponding medical record type into a teacher model for multiple rounds of abstract prompt word rewriting processing to obtain the candidate abstract prompt words for the corresponding medical record type; the first natural language description information is used to instruct the teacher model to perform the abstract prompt word rewriting processing; Inputting the optimized artificial abstract prompt words for each of the medical record types and the candidate abstract prompt words for the corresponding medical record type into the empty set of candidate abstract prompt words for the corresponding medical record type to obtain the candidate abstract prompt word set for the corresponding medical record type.

3. The method according to claim 2, wherein The obtaining the optimized artificial abstract prompt words for each of the medical record types includes: In response to receiving the initial artificial abstract prompt words for each of the medical record types and the original medical records for the corresponding medical record types, inputting each of the initial artificial abstract prompt words and the original medical records for the corresponding medical record types into the teacher model for medical record automatic abstract processing to obtain the medical record abstracts for the corresponding medical record types; In response to performing a first prompt optimization operation on the initial artificial summary prompt words of the corresponding medical record type according to the medical record abstracts of the respective medical record types, determine the optimized artificial summary prompt words of the corresponding medical record type according to the result of the first prompt optimization operation.

4. The method according to claim 1, characterized in that, The obtaining of the respective medical record types, the corresponding candidate summary prompt word sets of the medical record types, and the corresponding candidate scoring prompt word sets of the medical record types further includes: Obtain the optimized artificial scoring prompt words of each of the medical record types; In response to receiving the second natural language description information of each of the medical record types, input the optimized artificial scoring prompt words of each and the second natural language description information of the corresponding medical record type into the teacher model for multiple rounds of scoring prompt word rewriting processing to obtain the candidate scoring prompt words of the corresponding medical record type; the second natural language description information is used to instruct the teacher model to perform the scoring prompt word rewriting processing; Input the optimized artificial scoring prompt words of each of the medical record types and the candidate scoring prompt words of the corresponding medical record type into the empty set of candidate scoring prompt words of the corresponding medical record type to obtain the candidate scoring prompt word sets of the corresponding medical record types.

5. The method according to claim 4, wherein The obtaining of the optimized artificial scoring prompt words of each of the medical record types includes: In response to receiving the initial artificial evaluation prompt words of each of the medical record types, the original medical records of the corresponding medical record types, and the medical record abstracts of the corresponding medical record types, input the initial artificial evaluation prompt words of each, the original medical records of the corresponding medical record types, and the medical record abstracts of the corresponding medical record types into the teacher model for abstract scoring processing to obtain the first abstract scoring results of the corresponding medical record types; In response to performing a second prompt optimization operation on the initial artificial evaluation prompt words of the corresponding medical record type according to the first abstract scoring results of the respective medical record types, determine the optimized artificial scoring prompt words of the corresponding medical record type according to the result of the second prompt optimization operation.

6. The method according to claim 1, characterized in that, The performing of the abstract effect evaluation processing on each of the candidate summary prompt word sets and each of the candidate scoring prompt word sets to obtain the optimal summary prompt words of the corresponding medical record types and the optimal evaluation prompt words of the corresponding medical record types includes: Successively perform prompt word random extraction processing on the candidate summary prompt word sets of each of the medical record types to obtain the target candidate summary prompt words of the corresponding medical record types; Obtain each target teacher model and the original medical records corresponding to each of the medical record types; the target teacher model is a teacher model configured and processed according to each preset model algebra and each model parameter set; The original medical records corresponding to each of the medical record types, the target candidate abstract prompt words corresponding to the corresponding medical record types, and the candidate scoring prompt words in the corresponding candidate scoring prompt word sets corresponding to the corresponding medical record types are respectively input into each of the target teacher models to obtain the target teacher model abstract scoring result sets of the candidate scoring prompt words in the corresponding candidate scoring prompt word sets corresponding to the corresponding medical record types; the target teacher model set includes the second abstract scoring results of the corresponding target teacher models; Based on the target teacher model abstract scoring result sets of the candidate scoring prompt words in the corresponding candidate scoring prompt word sets of each of the medical record types and a preset standard abstract scoring result set, an analysis and comparison are performed to obtain the optimal evaluation prompt words corresponding to the corresponding medical record types.

7. The method according to claim 6, characterized in that, The performing of the abstract effect evaluation processing on each of the candidate abstract prompt word sets and each of the candidate scoring prompt word sets to obtain the optimal abstract prompt words corresponding to the corresponding medical record types and the optimal evaluation prompt words corresponding to the corresponding medical record types includes: The original medical records corresponding to each of the medical record types, the optimal evaluation prompt words corresponding to the corresponding medical record types, and the candidate abstract prompt words in the corresponding candidate abstract prompt word sets corresponding to the corresponding medical record types are respectively input into the teacher model to obtain the third abstract scoring results of the candidate abstract prompt words in the corresponding candidate abstract prompt word sets corresponding to the corresponding medical record types; The optimal abstract prompt words corresponding to the corresponding medical record types are determined according to the maximum values of the third abstract scoring results of the candidate abstract prompt words in the corresponding candidate abstract prompt word sets of each of the medical record types.

8. A device for generating an automatic abstract model of medical records, characterized in that, The device includes: A prompt word set acquisition module, configured to acquire each medical record type, the corresponding candidate abstract prompt word set corresponding to the medical record type, and the corresponding candidate scoring prompt word set corresponding to the medical record type; wherein, the medical record type includes outpatient and emergency medical records or inpatient medical records; the candidate abstract prompt word set includes each candidate abstract prompt word; the candidate scoring prompt word set includes each candidate scoring prompt word; An abstract effect evaluation module, configured to perform abstract effect evaluation processing on each of the candidate abstract prompt word sets and each of the candidate scoring prompt word sets to obtain the optimal abstract prompt words corresponding to the corresponding medical record types and the optimal evaluation prompt words corresponding to the corresponding medical record types; An optimal student model determination module, configured to perform a test process on each candidate student model according to each of the optimal abstract prompt words and each of the optimal evaluation prompt words to obtain the test results of the corresponding candidate student models, and determine the optimal student model according to each of the test results; the optimal student model is the candidate student model corresponding to the optimal test result; A model training module, configured to obtain a training medical record sample set, train the optimal student model according to the training medical record sample set, each of the optimal summary prompt words, and each of the optimal evaluation prompt words, adjust the model parameters of the optimal student model based on the evaluation score indexes corresponding to the optimal evaluation prompt words obtained from training and testing until the evaluation scores of each of the optimal evaluation prompt words meet the preset requirements, and generate the trained medical record automatic summarization model, so as to output an optimal medical record summary based on the trained medical record automatic summarization model.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images